De-Biasing Generative Models using Counterfactual Methods

Download

• Paper

Abstract

Causal Counterfactual Generative Model (CCGM) is a VAE-based framework with a partially trainable causal layer that learns causal relationships without compromising reconstruction fidelity, allowing for bias analysis, interventions, and scenario simulations. Our method combines a causal latent space VAE model with modifications for causal fidelity, generating de-biased datasets from biased training data and offering finer control over the causal layer, as demonstrated by high-fidelity image and tabular data generation aligned with the causal framework.

Citation

Bhat, Sunay, Jeffrey Jiang, Omead Pooladzandi, and Gregory Pottie. 2022. “De-biasing Generative Models using Counterfactual Methods.” arXiv preprint arXiv:2207.01575.

@article{bhat2022debias,
  title={De-biasing Generative Models using Counterfactual Methods},
  author={Bhat, Sunay and Jiang, Jeffrey and Pooladzandi, Omead and Pottie, Gregory},
  journal={arXiv preprint arXiv:2207.01575},
  year={2022}
}